ZXT 1
Fast & lightQuick replies, small footprint. For chat, drafts and cleanups.
zxt-1✦ Reference · ZXT AI · Updated 8 November 2026
Three text assistants, one that sees photos, and an image maker. Pick a model by how fast you need the answer, how hard the question is, and whether ZXT needs to look at a picture.
Quick replies, small footprint. For chat, drafts and cleanups.
zxt-1The everyday model. Good reasoning at a reasonable speed.
zxt-2Slower, heavier, and the strongest at multi-step problems. Always thinks first.
zxt-3Reads photos, screenshots, receipts, handwriting and charts — then answers like ZXT 2.
zxt-v2Describe what you want — a photo, painting, logo, poster, wallpaper or icon — in any language. ZXT writes a detailed prompt, paints an original image and shows it above its reply with Download and Open buttons. Works from every chat model.
| Attribute | ZXT 1 | ZXT 2 | ZXT 3 | ZXT V2 |
|---|---|---|---|---|
| Best for | Chat, drafts | Daily work | Hard problems | Photos & documents |
| Speed | Fastest | Balanced | Slowest | Balanced |
| Context | 64k | 128k | 200k | 128k |
| Thinking | Off | Low · High | Low · Med · High | Off · High |
| Vision | — | — | — | Up to 4 photos |
| Create images | Yes | Yes | Yes | Yes |
| Web search | Yes | Yes | Yes | Yes |
| Files | 4 × 100 KB | 4 × 100 KB | 4 × 100 KB | 4 × 100 KB |
| API id | zxt-1 | zxt-2 | zxt-3 | zxt-v2 |
Image creation is handled by ZXT Image and works from every chat model. Speed and context figures assume the reference build; smaller machines reduce context first, then speed.
Thinking lets a model reason before it answers. The reasoning appears in a collapsible step above the answer — labelled “Thought for 12s” or similar — so you can open it or ignore it.
In the app, turn on Think beside the message box. ZXT 2 and ZXT V2 think only when it’s on, ZXT 3 always thinks, and ZXT 1 never does.
A short internal draft. Good for routine questions where speed matters.
Lets the model plan a couple of steps before answering.
Extended reasoning. Slower, but better at maths, logic and multi-part prompts.
Web sources are cited in the answer, never inside the reasoning step.
Turn on Search beside the message box. ZXT searches only when a question needs fresh information — greetings, small talk, coding and writing tasks never trigger a search.
Every site it read is grouped into one expandable Searched the web step, and the sources are listed with the answer. Works with every model; available in the app, not through the API.
Attach up to 4 text files per message, up to 100 KB each, from the + menu. Plain text, Markdown, CSV, JSON, source code and similar formats work with every model.
Choose Add photos from the + menu, or drop pictures on the message box: up to 4 photos per message (JPG, PNG, WebP). Large photos are resized to 1600 px before upload.
Only ZXT V2 can read photos. If another model is active, attaching a photo switches to ZXT V2 for you; the text models never get the photo option.
Just ask — “make a picture of a lighthouse at dusk” or “buatkan gambar kucing astronot” — or pick Create image from the + menu (or type /@). ZXT Image paints an original picture in a few seconds.
Formats: square 1024 × 1024, portrait 832 × 1248, landscape 1248 × 832. Each image has Download, Open and Copy prompt buttons. Requests that break the safety rules are declined.
Something not working — an error, a missing chat, trouble signing in? Email ZXT support. Describe what happened and what you expected, and add a screenshot if you can.